I have some questions related to the benchmarking numbers published at RTI site.
Related to one-to-one throughput numbers, it shows a single publisher thread sending around 450K messages per second on a 1G network, when i try to reproduce this at my end per publisher thread number is quite low.. and the reason i see is that the CPU core is completely utilized for my program thread. So i was wondering how did the benchmarking results show such a high number.
i even tried running the performance test under JAVA/performance/throughput but that too gives me a lower number, no-where near to what is published. And looking at the CPU usage at my end the output that i get seems rational. But the benchmarking results show something which is quite different..
Any thoughts on this.
We have a video with tips on running the RTI performance test: http://vimeo.com/user14186439/review/63789983/f347feece2
For example, the value you use for the batchSize command line argument can have a big affect on throughput.
Ok thanks. Does it mean the performance numbers published are with batch mode on. My application need is to NOT use batching and then capture the throughput. So it seems that the numbers that i get are maybe right if batching is turned off. Has anyone done benchmarks without batching.